How to analyse a spreadsheet with AI and check the numbers

Guides · 4 min read · Written by the Taufa team · Updated 30 October 2026

Most people who work with spreadsheets are not data analysts. They have a sales sheet, an expense file or a set of survey responses and a question: what sold best, where did the money go, what do customers say. An assistant that can read the file removes much of the fiddly work. The danger is a wrong number that looks right, so this guide pairs each step with a way to check it.

Prepare the file before you ask anything

Tidy data gives better answers. A few minutes of preparation saves a lot of confusion.

• Put column names in the first row, one per column, with clear names such as "Order date" and "Amount (INR)".
• Keep one table per sheet. Remove blank rows, merged cells and notes typed in the middle of the data.
• Make sure dates are real dates and money is a number, not text like "Rs. 1,200/-".
• Remove or mask personal details you do not need, such as phone numbers or addresses.

Start by asking what is in the file

Before analysing, ask the assistant to describe the data: how many rows, what each column holds, how many values are missing, and anything odd. This catches problems early, such as a column that is half empty or two different spellings of the same city.

Clean it, and keep a record

Ask for cleaning steps one by one, and ask the assistant to tell you what it changed. Then keep the original file untouched and save the cleaned version separately, so you can always go back.

• "Standardise the city names (Bengaluru, Bangalore, BLR) and list every change you made."
• "Remove exact duplicate rows and tell me how many you removed."
• "Convert the Amount column to numbers and show me any rows where this failed."

Ask specific questions

"Analyse this" gets a generic reply. A specific question gets a useful one. Say what you want to decide.

Where the assistant can run code on your file, as Taufa can with Python, it calculates the answer instead of estimating it from the text it can see. That is usually more reliable for totals, averages and filters. Ask it to show the calculation or the formula it used, so you can see how it reached each figure.

How to check the numbers

1. Check the grand total yourself. Sum the column in your spreadsheet and compare it with the assistant's figure. If the totals differ, stop and find out why.
2. Spot-check three rows. Pick a few entries, find them in the file and confirm they were counted as described.
3. Look at the count. If you asked about 2,000 orders and the answer is based on 1,874, ask which rows were left out and why.
4. Ask for the formula or code, and read it. Look for filters you did not ask for, such as dates cut off or a category excluded.
5. Sense-check the result. If a month shows ten times its usual sales, there is either a story or an error.

Charts that help

Ask for the simplest chart that answers your question: lines for change over time, bars for comparing categories, and a table when exact figures matter. Ask it to label the axes with units and to start bar charts at zero. If a chart looks dramatic, check the scale before believing it.

Turn the result into something to share

Once you trust the numbers, ask for a short written summary for the person who will read it, in plain language: the answer first, then the evidence, then caveats such as missing data or a small sample. If your assistant can create files, you can ask for the tables as a new Excel sheet or the summary as a Word document or slides.

Related guides

• How to check an AI answer before you rely on it
• Learn Python with an AI assistant: a beginner's plan
• How to summarise a long PDF with AI and trust the result

AI can be wrong, so check anything that matters. About Taufa · Start free